Last time, we revised the parameters of the partially observable world:
The goal becomes a percept when it becomes visible. The robot is teleported back to a vertex. This adds a couple constraints, to be sure; but we're optimizing for distance: coming up with a balance of exploration vs. goal-directed traversal is still an interesting problem.
Next time, furthermore, when we add non-determinism (i.e. the robot won't necessarily arrive at the expected destination); things will get more interesting.